Pollen sterols are associated with phylogenetics and environment but not with pollinators

Phytosterols are primary plant metabolites that have fundamental structural and regulatory functions. They are also essential nutrients for phytophagous insects, including pollinators, that cannot synthesize sterols. Despite the well-described composition and diversity in vegetative plant tissues, few studies have examined phytosterol diversity in pollen. We quantified 25 pollen phytosterols in 122 plant species (105 genera, 51 families) to determine their composition and diversity across plant taxa. We searched literature and databases for plant phylogeny, environmental conditions, and pollinator guilds of the species to examine the relationships with pollen sterols. 24-methylenecholesterol, sitosterol and isofucosterol were the most common and abundant pollen sterols. We found phylogenetic clustering of twelve individual sterols, total sterol content and sterol diversity, and of sterol groupings that reflect their underlying biosynthesis pathway (24 carbon alkylation, ring B desaturation). Plants originating in tropical-like climates (higher mean annual temperature, lower temperature seasonality, higher precipitation in wettest quarter) were more likely to record higher pollen sterol content. However, pollen sterol composition and content showed no clear relationship with pollinator guilds. Our study is the first to show that pollen sterol diversity is phylogenetically clustered and that pollen sterol content may adapt to environmental conditions.


Summary 18
• Phytosterols are primary plant metabolites that have fundamental structural and 19 regulatory functions. They are also essential nutrients for phytophagous insects, 20 including pollinators, that cannot synthesize sterols. Despite the well-described 21 composition and diversity in vegetative plant tissues, few studies have examined 22 phytosterol diversity in pollen. 23 • We quantified 25 pollen phytosterols in 122 plant species (105 genera, 51 families) 24 to determine their composition and diversity across plant taxa. We searched 25 literature and databases for plant phylogeny, environmental conditions, and 26 pollinator guilds of the species to examine the relationships with pollen sterols. 27 • 24-methylenecholesterol, sitosterol and isofucosterol were the most common and 28 abundant pollen sterols. We found phylogenetic clustering of twelve individual 29 sterols, total sterol content and sterol diversity, and of sterol groupings that reflect 30 their underlying biosynthesis pathway (24 carbon alkylation, ring B desaturation). 31 Plants originating in tropical-like climates (higher mean annual temperature, lower 32 temperature seasonality, higher precipitation in wettest quarter) were more likely 33 to record higher pollen sterol content. However, pollen sterol composition and 34 content showed no clear relationship with pollinator guilds. Royal Botanic Gardens (RBG), Kew, UK and nearby areas (see Table S1 for details 111 of collection dates and locations for each species). RBG Kew supports a diverse 112 collection of living plant species from across the world. Prior to pollen collection, we 113 used a fine-meshed bag to cover flower buds whenever possible to prevent potential 114 contamination or removal due to pollinator visitation. When flowers were fully open, 115 we gently shook the flower and collected pollen into a weighed 2 mL microcentrifuge 116 tube (Eppendorf®, Safe-Lock TM ). For species for which pollen was more difficult to 117 harvest, such as in the cases of Lamium purpureum L. and Ulex europaeus L., we 118 used small forceps to help push the pollen out or trigger pollen ejection, respectively. 119 Pilot studies carried out in our laboratory (with Helleborus foetidus, Prunus avium, 120 Prunus spinosa, Salix cinerea and Symphytum officinale) showed a conserved pattern 121 of pollen sterol composition: within species variation was significantly lower than 122 between species variation (all p-values < 0.001 under multi-variate distribution tests 123 e.g., Hotelling test, Pillai test, and Wilks' lambda distribution test), consistent with 124 findings on within vs. between species variation of other pollen metabolites (Palmer-125 Young et al., 2019). Therefore, we collected 2 to 5 replicates per species (details see 126 Table S1) and used the average quantities across replicates of each species for 127 analyses. In total, we collected 308 samples from 122 species, representing 105 128 genera, 51 families and 25 orders across the major groups of seed plants 129 (Gymnosperms, Nymphaeales, Monocots, Ranunculids, Caryophyllales, Asterids and 130 Rosids; Table S1). Our selection of species was guided by a combination of practical 131 considerations (feasibility to collect sufficient pollen for analysis, availability of species 132 at Kew) while attempting to maximize phylogenetic and ecological diversity of plants 133 (pollinator guilds, ecological niches). Pollen weight (to 0.1 mg accuracy) and collection 134 date were recorded for each sample. Pollen samples were stored in a freezer (-20°C) 135 before extracting sterols. 136

Sterol content analysis 137
To extract sterols and stanols (from here referred to as phytosterols or pollen sterols) 138 from the pollen, we added 500 μl 10% KOH in MeOH to the microcentrifuge tubes 139 containing a weighed pollen sample. Then, an internal standard (20 μl of 0.5 mg ml -1 140 epicoprostanol) was added prior to incubating the tube for 2 h at 80°C for 141 saponification. Phytosterols were then recovered into 1 mL hexane. After complete 142 evaporation of hexane, phytosterols remained in the tube. We derivatized these with 143 20 μl Tri-Sil (Sigma, Gillingham, Dorset UK) and then briefly vortexed and injected 144 directly into an Agilent Technologies (Palo Alto, CA, USA) 7890A gas chromatograph 145 connected to an Agilent Technologies 5975C MSD mass spectrometer (GC-MS) and 146 eluted over an Agilent DB5 column using a splitless injection at 250°C with a standard 147 GC program at 170°C for 1 minute ramped to 280°C at 20°C per minute and monitoring 148 between 50 and 550 amu. 149 All 25 phytosterols were identified by comparison of their retention time relative to 150 cholesterol and mass spectra from authentic standards (David W Nes collection, 151 details see Fig. S4 for mass spectra of each sterol) either directly through co-analysis 152 or using existing data and confirmed where data was available with the NIST (National 153 Institute of Standards and Technology) mass spectral library (Guo et al., 1995;Heupel 154 and Nes, 1984;Nes et al., 1977;Xu et al., 1988;Zhou et al., 2009;Nes et al., 2003). 155 To quantify the amount of each phytosterol, we used its relative peak area by 156 calculating the ratio of the peak area of the targeted sterol to that of the internal 157 standard. Then, by multiplying the ratio with the mass of the internal standard, we 158 obtained the mass of each sterol in the sample. Compound identification (using target 159 ion) and quantification were carried out with ChemStation Enhanced Data Analysis 160 (Version E.01.00). In total, we identified 25 phytosterols in pollen (Table S1). 161 For each plant species, we calculated each phytosterol amount (µg per mg sampled 162 pollen), total sterol content (µg per mg sampled pollen), and the percentage of each 163 sterol in total phytosterol content. In addition, we calculated the chemical diversity 164 index using Shannon entropy: where S is the total number of phytosterols, pi is the 165 percentage of the i th phytosterol. Note that we used the total phytosterol number S as 166 the base of log (instead of the natural base e) to scale the range of diversity index 167 values to [0, 1] with 1 indicating the highest diversity. This equates to calculating 168 Shannon's equitability. Finally, for each phytosterol, we calculated its commonness 169 and abundance across all plant species. Commonness is given by the proportion of 170 plant species that contained that specific phytosterol (i.e., present/absent). Relative 171 abundance was given by the average proportion of a specific sterol across all species. 172 Additionally, to understand how different phytosterol in pollen co-varied, we performed 173 a factor analysis using the R package stats (R Core Team, 2020) on the data for the 174 absolute weight of phytosterols measured in pollen across the entire data set. We set 175 a criterion of eigenvalue > 1 for inclusion of extracted factors. A varimax rotation was 176 used to adjust the fit of the factor analysis to variance in the data. 177 Moreover, based on biosynthetic reasoning as discussed by Benveniste (2004), we 178 arranged these phytosterols identified in our pollen samples into alternate hypothetical 179 biosynthetic pathways to cholesterol and 24-alkyl phytosterols. 180

Phylogenetic tree construction and analyses 181
We used the R package rotl (Michonneau et al., 2016) to download the induced 182 subtree of only our focal taxa from the Open Tree of Life (OTL) synthetic tree (Hinchliff 183 et al., 2015;Rees et al., 2017). If only the genus was known, OTL used the root of the 184 genus for the subtree wherever possible. Name synonyms and corrections suggested 185 by OTL for genus and species were adopted in our analyses (see Table S2). Taxa  To determine whether there is phylogenetic structure in the pollen sterol data, we used 204 the function phyloSignal from the R package phylosignal (Keck et al., 2016)  records from species reported outside of their native regions at the level-2 (regional or 237 sub-continental level). Finally, we removed duplicates and thinned each species' 238 occurrence dataset by keeping only one record by 10x10 arc-min grid cell to limit 239 spatial autocorrelation. In total, 355,912 occurrence records were retrieved across all 240 species (Table S1).  Table S3). soil pH and negatively with precipitation, and PC3 being positively correlated with soil 265 depth to bedrock and negatively with land slope (see Table S3 for variable 266 contributions to PCA axes). To quantify the niche breadth of each species, we first 267  Table S3). Because three-dimensional alpha shapes require at least 278 five occurrence points to be drawn, species with fewer records were discarded. We 279 also discarded those species lacking sufficient and reliable geographic data or 280 taxonomic uncertainty (i.e., we did not extract occurrence records for genera, 281 subspecies and hybrids). In the end, we quantified niche breadth for 90 species, while 282 32 taxa were discarded, and niche position for 100 species (22 taxa discarded; details 283 see Table S1). 284

Pollinator data collection 285
To study whether there is a relationship between plants' pollen sterols and their 286 pollinators, we categorized plants in two different ways. Firstly, based on pollinator 287 guilds, as 1) Bee, 2) Fly, 3) Lepidoptera, 4) Thrips, 5) Generalist insect, 6) Bird, or 7) 288 Wind pollinated. Secondly, we grouped plants by whether or not pollen acts as a 289 reward for bee pollinators. On the one hand, bees depend on pollen as larval food and 290 require pollen sterols as essential nutrients. Plants could therefore hypothetically 291 attract bee pollinators with pollen sterol profiles of high nutritional quality to them. On 292 the other hand, if pollen does not play a role as bee reward (i.e., in non-bee pollinated 293 plants, and/or where nectar is the sole reward), sterol profiles could be expected that 294 are of low quality or even toxic to bees to prevent pollen robbery (as shown for some 295 other chemical compounds in pollen, Rivest & Forrest 2019). 296 To classify plant pollinator guilds and groups, we conducted literature searches for 297 each plant species on Google Scholar, using the scientific name (including common 298 synonyms) and "pollinat*", OR "pollen", OR "flower" as search terms. We examined 299 relevant cited or citing references of publications found in this way for additional 300 records, and consulted Knuth (1908Knuth ( , 1909 and Westrich (2018), or personal 301 observations. If no sources on pollination and flower visitation were available, the 302 pollinator guild was classified as "unknown" (10 species in data set). We included 303 plant species in the "pollen as bee reward" group that both receive pollination services 304 by bees (including some plants in the "generalist insect pollination" category) and have 305 records of bees collecting pollen. Plants were classified as not producing pollen as 306 bee reward if they were either not pollinated by bees, or, in case of bee pollination, 307 had clear evidence of pollen not being collected by bees (e.g., pollen contained in 308 pollinia of bee-pollinated orchids). Plants for which data on pollinator guild and 309 collection of pollen by bees was missing were classified as "unknown" (34 species in 310 data set). A full list of relevant references and the assigned pollinator guilds is provided 311 in Table S1. 312

Covariance of pollen sterols 387
The factor analysis reduced the data to 12 independent latent factors that explained 388 73% of sterol variation (Table 1). Overall, phytosterols that have close positions in their 389 biosynthetic pathways (Fig. 3) or use the same enzyme (e.g., reductase) for production 390 tend to align together with the same factors. For example, iso-obtusifoliol is the 391 precursor of 24-methylenelophenol, then it branches to either epifungisterol or 392 avenasterol via episterol (Fig. 3). These four sterol compounds (not including 393 episterol) largely aligned together with factor 1 which accounted for ~9% of the 394 variance (Table 1). Similar patterns also applied to factor 3 and factor 4 whose main 395 contributing sterols represented the early cyclopropyl pathway intermediates. Factor 396 5 represented a strong positive correlation among the stanols (saturated in ring B), 397 campestanol and sitostanol. Factors 6 and 7 represent products of Δ-24 reduction. In 398 addition, we found one inverse relationship between four of the most common 399 phytosterols (in factor 2, accounting for 8% of the variance), where 24-400 methylenecholesterol was aligned in the opposite direction as the presence of three 401 other phytosterols: sitosterol, campesterol and stigmasterol. 402

Phylogenetic patterns 403
We found significant phylogenetic signal in 12 out of 25 phytosterols (percentages of 404 individual compounds), of which 7 were significant for both Pagel's λ and Blomberg's 405 K, and 5 for only one of the tests (Fig.2, Table 2). When grouping phytosterols based 406 on the substitution at C-24 (24C-methyl-, 24C-ethyl-, or 24C-0) or based on the 407 position of methine in ring B (Δ 0 , Δ 5 , Δ 7 , Δ 8 ), we found a significant phylogenetic signal 408 (both Pagel's λ and Blomberg's K) for all groups except the Δ 8 sterols (Fig.2, Table 2). 409 Additionally, we found a significant signal for the Shannon diversity index and total 410 sterol content (µg sterol per mg pollen; Fig.2, Table 2). These results remain largely 411 consistent when excluding all taxa which are only identified to genus level. Note that 412 λ and K are largely agreeing on which phytosterols showed significant signal (Table  413 2), although the significant estimates for λ are relatively high (0.585 to 1, mean = 0.  Table 3). 426 Total pollen sterol content of plant species was positively correlated with some of the 427 environmental variables in their native range, but in general the explained variance 428 (r 2 ) was low (Fig. 4, Table S4). Specifically, total sterol content correlated with 429 environmental PC1 (associated with high mean temperatures, low temperature 430 seasonality and low soil carbon content; p = 0.015, r 2 = 0.060; Fig. 4). For linear 431 models of individual environmental variables, species with higher total pollen sterol 432 content tended to occur in locations with higher annual mean temperature, higher 433 temperatures in the coldest quarter, higher precipitation in the wettest quarter, and 434 lower temperature seasonality (p-values < 0.05 for linear models of phylogenetic 435 independent contrasts, r 2 between 0.05 to 0.08, Table S4), as is the case in tropical 436 conditions. For Shannon's H diversity of pollen sterol profiles, the only significant 437 association with environmental variables was a weak negative correlation with 438 temperature seasonality (p = 0.014; r 2 = 0.06) ( Table S4). None of the other 439 environmental variables or principal components were significantly correlated with 440 sterol content or diversity, nor was the total environmental niche breadth (Fig. 4, Table  441 S4). 442

Sterols and pollinator guilds 443
We found overall pollen sterol profiles were largely overlapping between plant groups 444 with different pollinator guilds (bee, Lepidoptera, generalist insect, bird, unknown; Our factor analysis (Table 1) (Table 1)

Correlations of phytosterols with abiotic factors 521
The presence of different phytosterols could be evolutionary adaptations to 522 environmental conditions. We detected a positive relationship between sterol content 523 and temperature (particularly mean annual temperature and mean temperature of the 524 coldest quarter), and a negative correlation with temperature seasonality, even though 525 the overall association strength was low (Table S4). This indicates that plants found to us for sampling). Limited sampling towards extremes of the environmental gradients 539 may have reduced our scope to detect associations between abiotic factors and pollen 540 sterol characteristics. Future work should therefore be targeted at sampling additional 541 plant species of more extreme environments to fill this gap. Note that our species were 542 sampled at glasshouses (e.g., tropical glasshouse, alpine glasshouse) or outdoors at 543 Royal Botanic Gardens Kew and nearby areas (sampling details see Table S1) to get 544 a first estimate of pollen sterol diversity across a broad range of species. Future in-545 depth studies on how abiotic conditions affect pollen sterol variation within-species 546 deserve further attention to build a more complete overview of pollen sterol diversity 547 at different taxonomic levels. 548

Impact of sterol diversity on pollinators 549
Pollen sterol amount and composition did not differ significantly between bee 550 pollinated and non-bee pollinated plant species. This could indicate that pollen sterols 551 have generally not been under selection by bee pollinators although we acknowledge 552 that our analysis combined all bee pollinated plants into one group. Therefore, it 553 remains possible that pollen sterols play a role in finer scale interactions between 554 different bee species of varying levels of pollen specialization and their host plants. 555 We also note that, although we based our assessment of pollinator guilds on the best 556 available literature data, the quality of evidence for the effective pollinators of the 557 plants in our data set varied. This calls for further in-depth studies of the relationships 558 between pollen sterols and pollinators, also including wind-pollinated Angiosperm taxa 559 missing in this work as points of comparison to animal pollinated plants. where floral resources do not provision 24-methylenecholesterol. Our data suggested 572 that many Asteraceae (e.g., Achillea ptarmica L., Tanacetum vulgare L. Achillea 573 millefolium L., Jacobaea vulgaris Gaertn., Centaurea nigra L. and Cirsium vulgare 574 (Savi.) Ten) are rich in Δ 7 -sterols (Fig. 2, Table S1) and lack the common honeybee-575 favourable Δ 5 -sterols (e.g., 24-methylenecholesterol). Δ 7 -sterols are known to be toxic 576  linear models (with intercept set to zero); r 2 and p-values for linear models inserted in 975 the respective plot. PC loadings from each environmental variable see Table S3. 976       Table S1. Data table (plant species, scores for different environmental 1013 variables/principal components, pollinator guilds, sterol composition (relative & 1014 absolute amounts)). 1015 Table S2. Scientific name and family for all sampled species, along with suggested 1016 OTL synonyms (which were subsequently used) and taxon IDs; species excluded 1017 from the phylogeny are highlighted in grey; reason for exclusion due to issues in the 1018 data and/or the OTL taxonomy are indicated. 1019 Table S3. Variable contributions to axes of PCA of 13 environmental variables. 1020 Table S4. Test results: Linear models of phylogenetic independent contrasts (PICs) 1021 of total sterol amount/diversity against PICs of environmental variables and niche 1022 breadth. 1023         Table S3.